PhD researcher · Boise State University
Building efficient, dependable machine learning systems.
I work across efficient AI, computer vision, LLM inference, and research tooling, with an emphasis on reproducibility, deployment constraints, and evidence that survives scrutiny.
Efficient AIML SystemsComputer VisionQuantizationEdge InferenceResearch Engineering
Professional profiles
Research
Efficient AI designed with deployment in mind
My work connects training-time decisions with the arithmetic, memory, latency, and verification constraints that determine whether a model is useful outside a notebook.
2026 · First-author IEEE paper
Multiplier-Free LLM Linear Layers via Weights-Only Power-of-Two QAT
Power-of-two quantization-aware training for Transformer linear layers, evaluated on DistilGPT-2 and Llama-3.2-1B with perplexity, throughput, memory, and arithmetic-energy proxies.
- Peer-reviewed at IEEE ICAD 2026
- Replaces constrained multiplications with bounded shift-add operations
- Reports matched evaluation across model quality and deployment cost
Current research
SPARQ
Single-term power-of-two quantization and multiplier-free hardware acceleration for efficient neural network inference. Manuscript and FPGA-oriented evidence package in active development.
Research principles
Reproducibility before claims
Versioned configurations, leakage-aware evaluation, parity checks, multi-seed results, documented limitations, and deployment-aware benchmarks.
Selected work
Projects that show the full engineering loop
From data quality and experiments to deployment checks and operator-facing interfaces. Each repository includes documentation, reproducible workflows, and explicit limitations.
Efficient AI education2026
Po2QAT: Power-of-Two Quantization Algorithm
A classroom-ready PyTorch implementation applying the same quantization-aware training algorithm to a MobileNet-style CNN, TinyViT, and TinyGPT, with inspectable exports and reproducible metrics.
View repositoryAgentic ML tooling2026
LLM Agent for Experiment & Knowledge Workflows
A grounded agent workspace for querying experiment reports, comparing runs, generating cited summaries, routing tools, and retaining session memory over real ML artifacts.
View repositoryBenchmarking2026
Edge AI Benchmark Suite
Compares predictive quality with latency, throughput, memory, model size, FLOPs, and estimated energy under realistic edge constraints.
View repositoryFull-stack ML platform2026
ML Experiment Control Center
Launches experiments, tracks run metadata, compares results, browses artifacts, and exports summaries through a product-style workflow.
View repositoryDeployment verification2026
Model Export & Benchmarking Toolkit
Exports models, checks cross-runtime output parity, benchmarks inference, records failures, and generates deployment reports.
View repositoryPredictive maintenance2026
Early Warning & Drift Detection
Combines supervised models, anomaly scoring, uncertainty, SHAP analysis, and sensor-distribution drift monitoring on NASA C-MAPSS.
View repositoryExplore more projects
Efficient LLM Inference ToolkitQuantization, sparsity, low-rank approximation
Efficient & Robust CV BenchmarkAccuracy, efficiency, and corruption robustness
Dataset Quality & Leakage PlatformPre-training quality gates for multiple data types
Customer Churn & Revenue RiskClassification and decision-support analytics
Smart Resume StudioNext.js application with PDF and DOCX export
Cat Vocalization InterpreterCautious audio classification with model documentation
Publications
Peer-reviewed work and preprints
Publication status is stated explicitly. For the most current citation record, see Google Scholar or ORCID.
2026
Peer-reviewed conference · First author
Multiplier-Free LLM Linear Layers via Weights-Only Power-of-Two QAT
Ikteder Akhand Udoy and Omiya Hassan · IEEE ICAD 2026 · DOI: 10.1109/ICAD69378.2026.11609079
2026
Peer-reviewed journal
Attention-Driven Hybrid Deep Learning for Automated Alzheimer’s Disease Severity Assessment via MRI Neuroimaging
Mahimul Islam Nadim et al., including Ikteder A. Udoy · Emerging Science Journal, Vol. 10, No. 3
2025
Peer-reviewed conference · First author
Lightweight Binarized Neural Network for Real-Time Sleep Apnea Detection on Edge Hardware
Ikteder Akhand Udoy, Rokaiya Sharmin, Md. Maruf Hossain, Syed Kamrul Islam, and Omiya Hassan · IEEE MeMeA 2025
2025
Peer-reviewed journal
AI-Driven Technology in Heart Failure Detection and Diagnosis: A Review of the Advancement in Personalized Healthcare
Ikteder Akhand Udoy and Omiya Hassan · Symmetry 17(3), 469
2024
Preprint
A Survey on the Application of Generative Adversarial Networks in Cybersecurity
Md Mashrur Arifin et al., including Ikteder Akhand Udoy · arXiv:2407.08839
Background
Research, engineering, and technical coordination
2023 – Present
Graduate Research Assistant
Boise State University · LPiNS Lab
- Builds Python and PyTorch pipelines for preprocessing, training, evaluation, orchestration, and reporting.
- Develops controlled experiments across classification, vision, and language workloads.
- Creates reusable validation, logging, checkpoint analysis, and model-comparison workflows.
Jan 2022 – Jun 2023
Software Engineer I
Playense · Software and Application Development
- Developed, tested, and debugged application features using Python, C++, and Dart.
- Implemented and integrated software components while investigating technical issues.
- Completed and delivered eight client projects through final handoff.
2023 – Present
Teaching Assistant
Boise State University
Supports algorithms, data structures, digital systems, AI hardware systems, programming, debugging, and problem solving.
PhD in Computing · Expected Dec 2027
Boise State University
Data Science and Machine Learning emphasis with research spanning computer vision, efficient AI, and hardware-aware machine learning. Advisor: Dr. Omiya Hassan.
BSc in CSE · 2017 – 2021
American International University-Bangladesh
Foundation in software engineering, algorithms, systems, and applied computing.
Capabilities
Tools used to turn research into reliable systems
ML & research
PyTorch, scikit-learn, computer vision, Transformers, QAT, model compression, error analysis, experimental design
Systems & deployment
ONNX Runtime, Linux, Docker, inference benchmarking, parity verification, edge evaluation, profiling
Software & data
Python, SQL, Java, C++, FastAPI, React, TypeScript, pandas, SQLite, Git and GitHub
Research practice
Reproducible workflows, dataset documentation, leakage controls, multi-seed evaluation, model cards, technical writing
Contact
Let’s build machine learning systems that hold up in practice.
I’m interested in research collaborations and software engineering, data science, applied scientist, ML systems, and R&D opportunities.